An in-house timer for ART-specific extra time in gynecologic pelvic online adaptive radiation therapy

Abstract

Online adaptive radiation therapy (ART) lengthens the session, but many published times report one interval and do not isolate ART-specific activity. We developed a console timer that records wall-clock timestamps for a defined event list, stores per-fraction ART settings, and exports Excel tables. ART-specific extra time extends from ART mode until plan approval. After verification on 100 simulated fractions, it was applied in 38 fan-beam CT-guided gynecologic pelvic online ART fractions (nine patients). Mean ART-specific extra time was 13.7 min within a 21.8-min session, dominated by optimization and target/organ-at-risk editing. It supports bottleneck audit under a shared event list.

Highlights

• •Console timer isolates ART-mode-to-approval extra time.
• •Used in 38 FBCT-guided gynecologic pelvic online ART fractions.
• •Mean ART-specific extra time was 13.7 min in a 21.8-min session.
• •Optimization plus CTV/OAR editing accounted for 71.1% of ART time.
• •Supports bottleneck audit; keypress capture remains a limitation.

Introduction

Online adaptive radiation therapy (ART) modifies the plan on the couch according to the anatomy of the day [1], [2]. In the gynecologic pelvis, targets move with bladder filling, uterine deformation, and rectal air or filling [3], [4]. The extra on-table work—auto-contouring, target and organ-at-risk (OAR) editing, plan generation, optimization, and approval—lengthens the session relative to image-guided radiation therapy (IGRT) and affects slot length and throughput [5]. Many published times are a single interval (door-to-door, cone-beam CT [CBCT] to beam-on, or “adaptation”) and remain dominated by prostate, mixed-site, or definitive cervix series [6], [7], [8], [9], [10], [11]. Other reports also partition the workflow into discrete steps, for example bladder [12], breast [13], [14], and same-day palliative adaptation [15], but endpoints, platforms, and staffing still differ. Manual clock-time logs rarely share one event list or export a table that can be merged across fractions.

We developed an in-house console timer for ART-specific extra time: wall-clock time from switching to ART mode until plan approval (Fig. 1A, blue steps, red bracket). Positioning, IGRT1, IGRT2, and delivery are recorded as auxiliary steps (Fig. 1A, gray) but excluded from that interval. Fig. 1B shows typical non-adaptive IGRT (Positioning–IGRT–Delivery) without the ART-mode interval. IGRT2 is usually absent, so ART-specific extra time is a lower bound versus IGRT. Per-fraction ART settings are stored with each export. We describe the software, a functional check on 100 simulated fractions, and first clinical use in fan-beam CT (FBCT)–guided gynecologic pelvic online ART.

Fig. 1.

Event list timed by the console timer. (A) Fourteen recorded timestamps are displayed as twelve boxes: Positioning and IGRT1 each combine a start and a complete stamp, followed by Switch to ART, Auto-contour, CTV edit, OAR edit, PTV expand, Plan create, Optimize, Plan approve, IGRT2, and Delivery. Gray boxes are auxiliary stages. Blue boxes are ART-related stages. The red bracket marks ART-specific extra time (ART mode through plan approval). (B) Typical non-adaptive IGRT session (Positioning–IGRT–Delivery) lacks the ART-mode interval and usually lacks IGRT2, so ART-specific extra time is a lower bound versus IGRT. ART, adaptive radiation therapy; IGRT, image-guided radiation therapy. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Materials and methods

Software

The application was written in Python 3 with a Tkinter interface. Four modules are integrated. Recording uses keyboard shortcuts (Page Down or Arrow Down for the next event, and Page Up or Arrow Up to undo) to store the wall-clock time of each event (hour:minute:second) at the treatment console. The software stamps the keypress and computes durations from those timestamps. It does not use a separate stopwatch. Reported durations are elapsed wall-clock intervals between consecutive stamps, not hands-on working time within a step.

Before timing, four ART settings are selected and exported with the duration table: scan range (5 cm superior and inferior to the target, or a larger field), slice thickness (3 mm or 5 mm), CTV generation (automatic contouring or registration-based), and plan generation (model-based or ART algorithm–based). These fields describe how the fraction was adapted and can change ART-specific extra time.

A single export writes the current fraction to Excel (duration sheet and event-timestamp sheet). Files in a records folder can be merged with bilingual (Chinese/English) headers. For each duration column the software reports sample size, mean, median, standard deviation, minimum, and maximum. Box plots and trend plots are exported at 300 dpi.

The clinical event list has 14 timestamps (8 ART-related and 6 auxiliary, Fig. 1): positioning start and complete, IGRT1 start and complete, switch to ART, CTV and OAR auto-contouring complete, CTV modification complete, OAR modification complete, planning target volume (PTV) expansion and auxiliary structures complete, plan creation and parameter review complete, optimization complete, plan evaluation and approval, IGRT2 complete, and treatment complete (delivery). Fig. 1A shows these 14 timestamps as 12 boxes because Positioning and IGRT1 each combine a start and a complete stamp. ART-specific extra time runs from switch to ART until approval.

An earlier 16-timestamp list split auto-contouring into three substages and was used for simulated files and two early clinical fractions. Those substages were summed so all clinical fractions share one table. The software is in-house research software, not a certified medical device (Chinese software copyright registration no. 2026SR0391704). Keypress recording can miss or delay a stamp. Automated console-log capture without operator keypresses is planned but not yet implemented.

Functional check

Before clinical use, 100 simulated fractions (20 virtual patients × 5 fractions) were generated with that earlier 16-timestamp list to test recording, Excel export, bilingual merge, statistics, visualization, and missing-data handling. The four modules completed successfully. Simulated durations are not clinical results.

Clinical application

After the functional check, the software was used during routine FBCT-guided online ART for gynecologic pelvic targets in the Department of Radiation Oncology, Peking University Third Hospital, on a CT-integrated linac (uRT-linac 506c, United Imaging). An operator at the console (typically the assigned medical physicist, with radiation therapists and a radiation oncologist for plan approval) recorded events concurrently with treatment. No additional imaging or treatment was performed for research. All reported timing data were de-identified before analysis. The study was approved by the Peking University Third Hospital Medical Science Research Ethics Committee (IRB00006761-M20250825). Written informed consent was waived for this de-identified workflow analysis.

Inclusion criteria were Excel exports from gynecologic pelvic FBCT-guided online ART fractions with a complete ART-mode-to-approval interval (4 December 2025–6 May 2026). Exclusion criteria were test or near-zero-duration timestamp files, duplicate exports, and the less complete file when two operators timed the same fraction. Diagnosis, age, technique, and prescription were obtained from the departmental oncology information system extract (23 June 2026) after identifiers were removed. An extract field that looked like disease stage actually stored treatment-course sequence numbers rather than FIGO stage, so FIGO stage is not reported.

Sixty-five raw workbooks included bilingual copies, dual-timer recordings, and inconsistent free-text identifiers. After removing exact duplicates, dual-timer pairs, one 1-s test file, and correcting one identifier error, 38 gynecologic pelvic fractions in nine patients remained. For two dual-timed sessions, discarded ART times differed by 6 s and 80 s (delayed ART-start keypress on the discarded file). All 38 fractions used scan range target ±5 cm, automatic CTV generation, and model-based plan generation. Reconstruction thickness was 3 mm in 13 fractions and 5 mm in 25.

Endpoints and statistics

The software check assessed whether simulated files could be recorded, merged, summarized, and plotted without failure. The clinical primary endpoint was ART-specific extra time. Secondary endpoints were total recorded session time (positioning start to beam-off) and step durations. We report means across timed fractions and, because fractions from the same patient are correlated, the mean of per-patient means (n = 9). Approximate 95% confidence intervals (CIs) used mean ± 1.96 × SE, with SE = SD/√n.

Implausible keypresses were identified by manual inspection during data curation, not by an automated software rule. A priori step thresholds flagged durations that are clinically implausible for an active console step rather than a formal estimate of user variability: positioning = 0 s (missed stamp), auto-contouring <15 s (insufficient for review of auto-contours), plan optimization <60 s, and plan creation >120 s (commonly reflecting a delayed “start optimization” stamp). These fractions were removed from the corresponding step mean only. ART time was also re-estimated after excluding the three short-contouring fractions. Dual-timer pairs provide a limited check of inter-operator stamp timing. Analyses used the software statistics module and were independently repeated in Python scripts.

Results

The 100 simulated files were recorded, merged, summarized, and plotted at 300 dpi without error.

Nine patients and 38 timed fractions were analyzed. Eight patients had postoperative endometrial carcinoma and one had postoperative uterine carcinosarcoma. All received FBCT-guided gynecologic pelvic volumetric modulated arc therapy online ART, 45 Gy in 25 fractions of 1.8 Gy, on the uRT-linac 506c. Mean age was 63 years (range 58–72). Timing coverage comprised five fractions in seven patients, two in one patient, and one in one patient.

Mean ART-specific extra time was 824 s (13.7 min, standard deviation [SD] 137 s, median 805 s, 95% CI 13.0–14.5 min, range 591–1146 s). Mean total recorded session time was 1310 s (21.8 min, SD 162 s, 95% CI 21.0–22.7 min, range 1062–1845 s). ART-specific extra time was 62.9% of mean total recorded session time. The mean of per-patient ART means was 13.9 min (95% CI 13.0–14.7).

Among ART-specific steps, mean plan optimization was 229 s (27.7% of ART-specific extra time), CTV modification 203 s (24.6%), and OAR modification 155 s (18.7%). Combined CTV/OAR modification was 43.3% of ART-specific extra time, and optimization plus CTV/OAR modification was 71.1%. Mean IGRT2 was 108 s (1.8 min) and was excluded from ART-specific extra time. Mean delivery was 187 s (SD 20 s). After excluding implausible keypresses from the corresponding step mean only, mean positioning was 59 s (n = 36), auto-contouring 85 s (n = 35), optimization 234 s (n = 37), and plan creation 30 s (n = 37). After excluding the three short-contouring fractions, mean ART-specific extra time was 13.9 min (n = 35). Step distributions are shown in Fig. 2, and the corresponding mean ± SD values are listed in Table 1.

Fig. 2.

Box plots of step durations for 38 timed fractions, in chronological order on the x-axis: Positioning, IGRT1, CTV & OARs Contouring, CTV Modification, OARs Modification, PTV Expansion, Plan Creation, Plan Optimization, Plan Evaluation, IGRT2, and Delivery. Blue boxes are ART-specific steps. Gray boxes are auxiliary steps. Red line, median. Green diamond, mean. Orange bracket, ART-specific extra time (mean 13.7 min). Gray bracket, total recorded session (mean 21.8 min). ART, adaptive radiation therapy; IGRT, image-guided radiation therapy; OAR, organ at risk; PTV, planning target volume. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 1.

Mean step durations for 38 timed gynecologic pelvic online ART fractions.

Step	Category	Mean ± SD (s)	Mean (min)
ART-specific extra time	Primary	824.4 ± 136.8	13.7
Total recorded session	Session	1310.1 ± 162.4	21.8
Plan optimization	ART	228.7 ± 61.9	3.8
CTV modification	ART	202.8 ± 81.6	3.4
OAR modification	ART	154.6 ± 75.5	2.6
Delivery	Auxiliary	186.7 ± 20.0	3.1
IGRT2	Auxiliary	108.1 ± 19.0	1.8
Auto-contouring	ART	78.4 ± 29.4	1.3
IGRT1	Auxiliary	81.6 ± 24.6	1.4
Plan evaluation	ART	67.4 ± 21.0	1.1
Positioning	Auxiliary	56.3 ± 31.3	0.9
PTV expansion	ART	54.2 ± 39.8	0.9
Plan creation	ART	35.1 ± 34.3	0.6

Values are mean ± SD. Category ART denotes ART-specific steps between ART mode and plan approval. Category Auxiliary denotes Positioning, IGRT1, IGRT2, and Delivery, which are recorded but excluded from ART-specific extra time. Primary row is ART mode to plan approval. Session row is Positioning start to beam-off. SD, standard deviation.

Discussion

This Technical Note reports an in-house timer and its first clinical use. The application timestamps a defined event list, computes ART-specific extra time from ART-mode switch to plan approval, stores per-fraction ART settings, and exports tables that can be merged without manual transcription. Intended uses are (i) local bottleneck identification (here optimization and CTV/OAR editing), (ii) longitudinal audit under a fixed event list, and (iii) multi-center comparison only when sites share event definitions, ART-setting vocabulary, and training. Many centers already time sessions informally or from vendor logs. The unmet need addressed here is a portable ART-mode-focused event list with one-click table export, independent of a single vendor console and complementary to system logs rather than a replacement for them.

FBCT-guided gynecologic pelvic online ART required approximately 14 min of ART-specific elapsed time and 22 min from positioning to beam-off. Table 2 summarizes published online ART times by study, system, operator model, endpoint, and reported duration, including the present series. The total session (21.8 min) is comparable to Jiang et al. (21.22 min) [10] and Sun et al. (22.82 min) [11] on CT-linac / FBCT pathways, and shorter than Shelley et al. (29 ± 5 min, CBCT to beam-on) [9]. Stepwise series [12], [13], [14], [15] show that single-interval summaries are not universal, but endpoints remain non-interchangeable. Adding mean IGRT2 (1.8 min) yields 15.5 min of ART-associated activity after the first image, excluding positioning, IGRT1, and delivery.

Table 2.

Published online ART timing reports by platform and staffing, with the present series for reference.

Study	System	Operator model	Endpoint	Reported time
Byrne 2022 [6]	Ethos (CBCT)	MDT	Recontouring + replanning	19 min (mean)
Stanley 2023 [7]	Ethos (CBCT)	MDT	Door-to-door / adaptive	≈35 / ≈20 min
Yock 2021 [8]	Ethos emulator	Emulator	Workflow (target edit)	24.4 (13.1) min
Shelley 2023 [9]	Ethos (CBCT)	MDT	CBCT to beam-on	29 ± 5 min
Jiang 2024 [10]	uRT CT-linac	MDT	Positioning to beam-off	21.22 min
Sun 2025 [11]	FBCT-linac	MDT	First image to delivery	22.82 min
Azzarouali 2023 [12]	Ethos (CBCT)	MDT	Session (stepwise)	median 30 min
Nelissen 2024 [13]	Ethos (CBCT)	RTT + RO/MPE	On-couch (breast)	median ≈13–14 min
Present study	uRT-linac 506c (FBCT)	Physicist-led MDT	ART mode to approval	13.7 ± 2.3 min

Columns list the imaging or treatment system, who operates the adaptive workflow, the published timing endpoint, and the reported duration. Endpoints and populations differ, so numerical comparison is limited. CBCT, cone-beam computed tomography; FBCT, fan-beam computed tomography; MDT, multidisciplinary team; MPE, medical physics expert; RO, radiation oncologist; RTT, radiation therapist.

Optimization and manual CTV/OAR editing predominated, consistent with cervix series in which target editing or auto-segmentation review occupies a large share of the session [8], [9], [11]. Combined CTV and OAR modification here was 6.0 min, shorter than the 13.1 min of cervix target editing in Yock et al.'s emulator study [8]. Reducing contour-editing workload and avoiding prolonged or repeated optimization remain priorities for process improvement.

Keypress timing is not equivalent to vendor system-log timing. During data curation, step durations were screened against the a priori thresholds listed in Section 2.4. Two dual-timer sessions differed by 6 s and 80 s in ART time, indicating variable operator stamp timing. Step minima of 0–20 s should not be interpreted as true process times. ART clock time from mode switch to approval was stable in the sensitivity analysis (13.7 vs 13.9 min). Future versions should prioritize automated event capture from the treatment console or DICOM workflow so that operators need not press keys during adaptation.

Several limitations follow. This single-center series included only 9 patients and 38 timed fractions on one CT-linac, whereas published timing papers often concern prostate, mixed-site, bladder, breast, palliative, or definitive cervix treatments [6], [7], [8], [9], [10], [11], [12], [13], [14], [15]. Three of four ART settings did not vary, timed fractions were not a complete consecutive series, and operator identity was not exported. Matched non-adaptive IGRT times were not collected, and keypress timestamps were not cross-checked against vendor logs. Simulated files show that the modules run but do not prove clinical-stamp accuracy. Multi-center comparison needs a locked event list, shared ART-setting vocabulary, and training.

Conclusions

This console timer records ART-specific extra time from ART mode to plan approval, with per-fraction ART settings and one-click export. In 38 FBCT-guided gynecologic pelvic online ART fractions, mean ART-specific extra time was 13.7 min (95% CI 13.0–14.5) within a 21.8-min session, mainly from optimization and CTV/OAR editing. Because IGRT2 lies outside that interval, the additional workload versus non-adaptive IGRT is about 14 min or more. Primary uses are local bottleneck identification and longitudinal audit under a shared event list. Multi-center comparison remains contingent on that list. Access is available from the corresponding author (registration no. 2026SR0391704).

Declaration of generative AI in scientific writing

During the preparation of this work the authors used a large language model (DeepSeek V3 web version) for English language editing and manuscript formatting. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

CRediT authorship contribution statement

Yongping Zheng: Writing – review & editing, Writing – original draft, Software, Methodology, Investigation, Data curation, Conceptualization. Runhong Lei: Writing – review & editing, Writing – original draft, Supervision, Methodology, Investigation, Data curation, Conceptualization. Ang Qu: Writing – original draft, Supervision, Funding acquisition, Conceptualization. Min Li: Writing – original draft, Validation, Data curation. Xiuwen Deng: Writing – original draft, Validation, Data curation. Haitao Sun: Writing – original draft, Data curation. Ping Jiang: Writing – original draft, Supervision, Conceptualization. Junjie Wang: Supervision, Conceptualization.

Ethics approval

The study was approved by the Peking University Third Hospital Medical Science Research Ethics Committee (IRB00006761-M20250825). The requirement for written informed consent was waived for this de-identified workflow analysis.

Ethics declaration

Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place.

This study was approved by the Peking University Third Hospital Medical Science Research Ethics Committee.

(Approval No. IRB00006761-M20250825)

Funding

This work was supported by Peking University Third Hospital (grant no. BYSYZHZB202522).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

De-identified duration tables are available from the corresponding author on reasonable request. The Adaptive Radiation Therapy (ART) Clinical Workflow Analysis System V1.0 is registered for Chinese software copyright (owner: Peking University Third Hospital, registration no. 2026SR0391704, certificate no. Soft. Copyright Reg. Cert. No. 17605985, registration date 6 March 2026). The timer can be obtained from the corresponding author for research and workflow improvement, subject to institutional agreement.